The rapid development of intelligent transportation systems (ITSs) has made vehicular networks (VNs) indispensable, particularly through vehicle-to-everything (V2X) communication. This study proposes an advanced framework for the construction and migration of digital twins (DTs) in vehicular networks to improve decision-making and predictive maintenance. The construction phase utilizes a large model-driven framework enhanced by an advanced deep reinforcement learning (DRL) algorithm, specifically an Enhanced Deep Q-Network (EDQN). This framework processes complex and dynamic vehicular data, supporting EDQN in optimizing decision-making processes. EDQN adapts dynamically to vehicular environments, ensuring high decision accuracy and efficiency. In the migration phase, due to limited base station coverage, transfer learning techniques are employed to enable the seamless migration of DTs across different base stations. This method minimizes computational overhead compared to traditional approaches by adapting pre-trained models to new environments with minimal retraining. Experimental simulations demonstrate that the integration of the large model architecture with EDQN significantly enhances decision-making processes. The transfer learning strategy effectively extends the operational coverage, maintaining high performance and service continuity during DT migration. This research underscores the potential of leveraging advanced AI techniques to improve the management and operational efficiency of vehicular networks, providing a robust foundation for future advancements in ITS.